arch/x86_64/kernel/setup64.c

   1 /*
   2  * X86-64 specific CPU setup.
   3  * Copyright (C) 1995  Linus Torvalds
   4  * Copyright 2001, 2002, 2003 SuSE Labs / Andi Kleen.
   5  * See setup.c for older changelog.
   6  * $Id: setup64.c,v 1.12 2002/03/21 10:09:17 ak Exp $
   7  */
   8 #include <linux/config.h>
   9 #include <linux/init.h>
  10 #include <linux/kernel.h>
  11 #include <linux/sched.h>
  12 #include <linux/string.h>
  13 #include <linux/bootmem.h>
  14 #include <asm/pda.h>
  15 #include <asm/pgtable.h>
  16 #include <asm/processor.h>
  17 #include <asm/desc.h>
  18 #include <asm/bitops.h>
  19 #include <asm/atomic.h>
  20 #include <asm/mmu_context.h>
  21 #include <asm/smp.h>
  22 #include <asm/i387.h>
  23 #include <asm/percpu.h>
  24 #include <asm/mtrr.h>
  25 #include <asm/proto.h>
  26 #include <asm/mman.h>
  27 #include <asm/numa.h>
  28
  29 char x86_boot_params[2048] __initdata = {0,};
  30
  31 unsigned long cpu_initialized __initdata = 0;
  32
  33 struct x8664_pda cpu_pda[NR_CPUS] __cacheline_aligned;
  34
  35 extern struct task_struct init_task;
  36
  37 extern unsigned char __per_cpu_start[], __per_cpu_end[];
  38
  39 extern struct desc_ptr cpu_gdt_descr[];
  40 struct desc_ptr idt_descr = { 256 * 16, (unsigned long) idt_table };
  41
  42 char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
  43
  44 unsigned long __supported_pte_mask = ~0UL;
  45 static int do_not_nx __initdata = 0;
  46 unsigned long vm_stack_flags = __VM_STACK_FLAGS;
  47 unsigned long vm_stack_flags32 = __VM_STACK_FLAGS;
  48 unsigned long vm_data_default_flags = __VM_DATA_DEFAULT_FLAGS;
  49 unsigned long vm_data_default_flags32 = __VM_DATA_DEFAULT_FLAGS;
  50 unsigned long vm_force_exec32 = PROT_EXEC;
  51
  52 /* noexec=on|off
  53 Control non executable mappings for 64bit processes.
  54
  55 on      Enable
  56 off     Disable
  57 noforce (default) Don't enable by default for heap/stack/data,
  58         but allow PROT_EXEC to be effective
  59
  60 */
  61 static int __init nonx_setup(char *str)
  62 {
  63         if (!strcmp(str, "on")) {
  64                 __supported_pte_mask |= _PAGE_NX;
  65                 do_not_nx = 0;
  66                 vm_data_default_flags &= ~VM_EXEC;
  67                 vm_stack_flags &= ~VM_EXEC;
  68         } else if (!strcmp(str, "noforce") || !strcmp(str, "off")) {
  69                 do_not_nx = (str[0] == 'o');
  70                 if (do_not_nx)
  71                         __supported_pte_mask &= ~_PAGE_NX;
  72                 vm_data_default_flags |= VM_EXEC;
  73                 vm_stack_flags |= VM_EXEC;
  74         }
  75         return 1;
  76 }
  77
  78 __setup("noexec=", nonx_setup);
  79
  80 /* noexec32=opt{,opt}
  81
  82 Control the no exec default for 32bit processes. Can be also overwritten
  83 per executable using ELF header flags (e.g. needed for the X server)
  84 Requires noexec=on or noexec=noforce to be effective.
  85
  86 Valid options:
  87    all,on    Heap,stack,data is non executable.
  88    off       (default) Heap,stack,data is executable
  89    stack     Stack is non executable, heap/data is.
  90    force     Don't imply PROT_EXEC for PROT_READ
  91    compat    (default) Imply PROT_EXEC for PROT_READ
  92
  93 */
  94  static int __init nonx32_setup(char *s)
  95  {
  96          while (*s) {
  97                 if (!strncmp(s, "all", 3) || !strncmp(s,"on",2)) {
  98                         vm_data_default_flags32 &= ~VM_EXEC;
  99                         vm_stack_flags32 &= ~VM_EXEC;
 100                 } else if (!strncmp(s, "off",3)) {
 101                         vm_data_default_flags32 |= VM_EXEC;
 102                         vm_stack_flags32 |= VM_EXEC;
 103                 } else if (!strncmp(s, "stack", 5)) {
 104                         vm_data_default_flags32 |= VM_EXEC;
 105                         vm_stack_flags32 &= ~VM_EXEC;
 106                 } else if (!strncmp(s, "force",5)) {
 107                         vm_force_exec32 = 0;
 108                 } else if (!strncmp(s, "compat",5)) {
 109                         vm_force_exec32 = PROT_EXEC;
 110                 }
 111                 s += strcspn(s, ",");
 112                 if (*s == ',')
 113                         ++s;
 114          }
 115          return 1;
 116 }
 117
 118 __setup("noexec32=", nonx32_setup);
 119
 120 /*
 121  * Great future plan:
 122  * Declare PDA itself and support (irqstack,tss,pml4) as per cpu data.
 123  * Always point %gs to its beginning
 124  */
 125 void __init setup_per_cpu_areas(void)
 126 {
 127         int i;
 128         unsigned long size;
 129
 130         /* Copy section for each CPU (we discard the original) */
 131         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 132 #ifdef CONFIG_MODULES
 133         if (size < PERCPU_ENOUGH_ROOM)
 134                 size = PERCPU_ENOUGH_ROOM;
 135 #endif
 136
 137         for (i = 0; i < NR_CPUS; i++) {
 138                 unsigned char *ptr;
 139                 /* If possible allocate on the node of the CPU.
 140                    In case it doesn't exist round-robin nodes. */
 141                 if (!NODE_DATA(i % numnodes)) {
 142                         printk("cpu with no node %d, numnodes %d\n", i, numnodes);
 143                         ptr = alloc_bootmem(size);
 144                 } else {
 145                         ptr = alloc_bootmem_node(NODE_DATA(i % numnodes), size);
 146                 }
 147                 if (!ptr)
 148                         panic("Cannot allocate cpu data for CPU %d\n", i);
 149                 cpu_pda[i].data_offset = ptr - __per_cpu_start;
 150                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 151         }
 152 }
 153
 154 void pda_init(int cpu)
 155 {
 156         pml4_t *level4;
 157         struct x8664_pda *pda = &cpu_pda[cpu];
 158
 159         /* Setup up data that may be needed in __get_free_pages early */
 160         asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
 161         wrmsrl(MSR_GS_BASE, cpu_pda + cpu);
 162
 163         pda->me = pda;
 164         pda->cpunumber = cpu;
 165         pda->irqcount = -1;
 166         pda->kernelstack =
 167                 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
 168         pda->active_mm = &init_mm;
 169         pda->mmu_state = 0;
 170
 171         if (cpu == 0) {
 172                 /* others are initialized in smpboot.c */
 173                 pda->pcurrent = &init_task;
 174                 pda->irqstackptr = boot_cpu_stack;
 175                 level4 = init_level4_pgt;
 176         } else {
 177                 level4 = (pml4_t *)__get_free_pages(GFP_ATOMIC, 0);
 178                 if (!level4)
 179                         panic("Cannot allocate top level page for cpu %d", cpu);
 180                 pda->irqstackptr = (char *)
 181                         __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
 182                 if (!pda->irqstackptr)
 183                         panic("cannot allocate irqstack for cpu %d", cpu);
 184         }
 185
 186         pda->level4_pgt = (unsigned long *)level4;
 187         if (level4 != init_level4_pgt)
 188                 memcpy(level4, &init_level4_pgt, PAGE_SIZE);
 189         set_pml4(level4 + 510, mk_kernel_pml4(__pa_symbol(boot_vmalloc_pgt)));
 190         asm volatile("movq %0,%%cr3" :: "r" (__pa(level4)));
 191
 192         pda->irqstackptr += IRQSTACKSIZE-64;
 193 }
 194
 195 char boot_exception_stacks[N_EXCEPTION_STACKS * EXCEPTION_STKSZ]
 196 __attribute__((section(".bss.page_aligned")));
 197
 198 /* May not be marked __init: used by software suspend */
 199 void syscall_init(void)
 200 {
 201         /*
 202          * LSTAR and STAR live in a bit strange symbiosis.
 203          * They both write to the same internal register. STAR allows to set CS/DS
 204          * but only a 32bit target. LSTAR sets the 64bit rip.
 205          */
 206         wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
 207         wrmsrl(MSR_LSTAR, system_call);
 208
 209 #ifdef CONFIG_IA32_EMULATION
 210         syscall32_cpu_init ();
 211 #endif
 212
 213         /* Flags to clear on syscall */
 214         wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000);
 215 }
 216
 217 void __init check_efer(void)
 218 {
 219         unsigned long efer;
 220
 221         rdmsrl(MSR_EFER, efer);
 222         if (!(efer & EFER_NX) || do_not_nx) {
 223                 __supported_pte_mask &= ~_PAGE_NX;
 224         }
 225 }
 226
 227 /*
 228  * cpu_init() initializes state that is per-CPU. Some data is already
 229  * initialized (naturally) in the bootstrap process, such as the GDT
 230  * and IDT. We reload them nevertheless, this function acts as a
 231  * 'CPU state barrier', nothing should get across.
 232  * A lot of state is already set up in PDA init.
 233  */
 234 void __init cpu_init (void)
 235 {
 236 #ifdef CONFIG_SMP
 237         int cpu = stack_smp_processor_id();
 238 #else
 239         int cpu = smp_processor_id();
 240 #endif
 241         struct tss_struct *t = &per_cpu(init_tss, cpu);
 242         unsigned long v;
 243         char *estacks = NULL;
 244         struct task_struct *me;
 245         int i;
 246
 247         /* CPU 0 is initialised in head64.c */
 248         if (cpu != 0) {
 249                 pda_init(cpu);
 250         } else
 251                 estacks = boot_exception_stacks;
 252
 253         me = current;
 254
 255         if (test_and_set_bit(cpu, &cpu_initialized))
 256                 panic("CPU#%d already initialized!\n", cpu);
 257
 258         printk("Initializing CPU#%d\n", cpu);
 259
 260                 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
 261
 262         /*
 263          * Initialize the per-CPU GDT with the boot GDT,
 264          * and set up the GDT descriptor:
 265          */
 266         if (cpu) {
 267                 memcpy(cpu_gdt_table[cpu], cpu_gdt_table[0], GDT_SIZE);
 268         }
 269
 270         cpu_gdt_descr[cpu].size = GDT_SIZE;
 271         cpu_gdt_descr[cpu].address = (unsigned long)cpu_gdt_table[cpu];
 272         __asm__ __volatile__("lgdt %0": "=m" (cpu_gdt_descr[cpu]));
 273         __asm__ __volatile__("lidt %0": "=m" (idt_descr));
 274
 275         memcpy(me->thread.tls_array, cpu_gdt_table[cpu], GDT_ENTRY_TLS_ENTRIES * 8);
 276
 277         /*
 278          * Delete NT
 279          */
 280
 281         asm volatile("pushfq ; popq %%rax ; btr $14,%%rax ; pushq %%rax ; popfq" ::: "eax");
 282
 283         if (cpu == 0)
 284                 early_identify_cpu(&boot_cpu_data);
 285
 286         syscall_init();
 287
 288         wrmsrl(MSR_FS_BASE, 0);
 289         wrmsrl(MSR_KERNEL_GS_BASE, 0);
 290         barrier();
 291
 292         check_efer();
 293
 294         /*
 295          * set up and load the per-CPU TSS
 296          */
 297         for (v = 0; v < N_EXCEPTION_STACKS; v++) {
 298                 if (cpu) {
 299                         estacks = (char *)__get_free_pages(GFP_ATOMIC,
 300                                                    EXCEPTION_STACK_ORDER);
 301                         if (!estacks)
 302                                 panic("Cannot allocate exception stack %ld %d\n",
 303                                       v, cpu);
 304                 }
 305                 estacks += EXCEPTION_STKSZ;
 306                 t->ist[v] = (unsigned long)estacks;
 307         }
 308
 309         t->io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
 310         /*
 311          * <= is required because the CPU will access up to
 312          * 8 bits beyond the end of the IO permission bitmap.
 313          */
 314         for (i = 0; i <= IO_BITMAP_LONGS; i++)
 315                 t->io_bitmap[i] = ~0UL;
 316
 317         atomic_inc(&init_mm.mm_count);
 318         me->active_mm = &init_mm;
 319         if (me->mm)
 320                 BUG();
 321         enter_lazy_tlb(&init_mm, me);
 322
 323         set_tss_desc(cpu, t);
 324         load_TR_desc();
 325         load_LDT(&init_mm.context);
 326
 327         /*
 328          * Clear all 6 debug registers:
 329          */
 330
 331         set_debug(0UL, 0);
 332         set_debug(0UL, 1);
 333         set_debug(0UL, 2);
 334         set_debug(0UL, 3);
 335         set_debug(0UL, 6);
 336         set_debug(0UL, 7);
 337
 338         fpu_init();
 339
 340 #ifdef CONFIG_NUMA
 341         numa_add_cpu(cpu);
 342 #endif
 343 }